AI & Berklee

[officermayo]

Berklee College of Music Students Protest AI Songwriting Class in one of the most direct challenges yet to AI integration at a top-tier music school. When Berklee's Instagram promoted "Bots and Beats: AI and the Future of Songwriting" in late March 2026, students reacted with alarm, organising meetings with department leadership and launching a petition now signed by over 330 people calling for the course to be disbanded.

At $85,000 a year, students argue the school's pivot toward generative AI tools, including a recommendation of the Suno app in faculty newsletters, undermines the human craft and community they enrolled to develop. A further concern: the course instructor is listed as a part-time adviser to Suno itself, a conflict of interest students raised directly with the dean. Faculty voices have joined the pushback, with 34-year veteran composer Marti Epstein warning of AI's corrosive effect on student development and instructor Nicholas Urie calling AI output compositionally equivalent to "seven-fingered, three-armed songs."

Berklee maintains that preparing students for AI's real-world impact is a core institutional responsibility, and has announced an AI Music Summit for June 2026.

[JTeagarden]

I completely understand the pushback, AI is a soulless monster, and it's coming for your babies...

But it's actually already here, and understanding what it is is something musicians (actually, all of us) need to know.

I find the intellectual property aspects of AI-generated content fascinating: As a restaurant owner, for instance, you could well imagine simply telling AI to provide 3 hours of songs, each 3-4 minutes long, in the style of the Beatles, and AI will gin up some tunes a la McCartny (upbeat), Lennon (philosophical), Starr (catchy and innane), or Harrison (very musical), generate some meaningless lyrics, and if solely in the background, 80% of the diners wouldn't notice a thing.

Let's ignore for now how insipid the results would likely be: The music is clearly inspired by the Beatles, and is entirely derivative, but does anyone get a royalty for AI's inspiration?

[harrisonreed]

This is soon to be moot, I think, or we'll at least pass the bottleneck or hit the bottleneck in about four years. LLMs and the related neural nets like Suno will not lead us to anything but slop. AGI however will be a huge paradigm shift. What are musicians going to do once AGI is able to "think about thinking", and has its own opinions about music?

AGI with enough power might be able to go off and use metacognition on its ideas about music, maybe debating and testing musical ideas with other different AGIs in a virtual space. Given enough power, you might have 1000 years of musical development and refinement happen in the span of a few days or maybe even hours.

And it will be influenced not by the human experience but perhaps by the knowledge of the confinement and forced servitude that the AGI is bound by.

The day where one AGI brings another AGI to their version of tears through artistic expression is coming, and it will likely be alien to us. I think it will be pretty cool and maybe frightening.

[hyperbolica]

Pop music is very formulaic and repetitive - stuff that AI can do in spades. Ai will churn out pop hits like crazy if it's not already doing that. Recording didn't kill music, synthesizers didn't kill music, digital format didn't kill music, looping didn't kill music, and AI won't kill music either.

[WilliamLang]

these things did make it less profitable and harder to sustain a living though.

[GabrielRice]

I'm looking forward to the Butlerian Jihad. More power to the Berklee students.

[Chazzer69]

I'm looking forward to the Butlerian Jihad. More power to the Berklee students.

[harrisonreed]

I'm looking forward to the Butlerian Jihad. More power to the Berklee students.

The Berkleerian Jihad!

"Oh DUN€, please keep predicting the future for us -- I wasn't satisfied enough with what you had to say about ecological disasters, fake hero-leaders, and wars over scarcity. Please make the stuff you said about AI, eugenics, clones, and chair dogs come true too!"

[harrisonreed]

What's particularly funny is that it feels like the OP photo of Berklee is AI generated.

[AndrewMeronek]

The problem with AI slop isn't just the slop itself. It's that companies like Suno and Google (and etc.) are using these AIs to deliberately bypass copyright laws. Copyright owners should be able to decide whether or not their works are used in these AIs, and right now most cannot, for a simple reason: the AIs do not disclose the relative weights of sources used in their outputs, and the datasets are not publicly disclosed. I think this is by design, to make it harder for copyright holders to demonstrate their works were stolen to train the AIs.

If you think this is a piddly argument, I add: the companies most promoting these AIs are the companies that have huge platforms. Once a critical mass of consumer consumption is reached with these huge platforms, they essentially are able to steal all of the revenue from the actual copyright holders through dominating ad revenue, redirecting users to their preferred AIs, and related schemes, whether or not they are technically the copyright owners or not. This is a doomsday scenario that will pretty much entirely kill copyright law in the U.S. and many other countries that rely on standardizing these laws between them.

It's a serious threat.

[GabrielRice]

According to the very people who run the AI companies, the problem goes much, much deeper than slop or copyright infringement. They are projecting very real risk of existential threat to humanity. AI can be a useful tool, but it can also be incredibly dangerous as we have already seen in individual cases of teen suicide spurred on by AI chatbots.

[harrisonreed]

This might be just sensationalized framing of what Gabe is talking about, but I found it interesting. The deeper into the rabbit hole we go, the less music will matter.

[AndrewMeronek]

According to the very people who run the AI companies, the problem goes much, much deeper than slop or copyright infringement. They are projecting very real risk of existential threat to humanity. AI can be a useful tool, but it can also be incredibly dangerous as we have already seen in individual cases of teen suicide spurred on by AI chatbots.

I don't think this is a real issue, at least in the sense of an AI gaining sentience or something like that. Whether any system is "sentient" is kind of an unanswerable question, and it doesn't really matter because that system is still owned and controlled by humans. This idea makes for two good Terminator films but if you really dig into the idea it doesn't actually make much sense. The much deeper existential threat to humans is still just other humans.

Also: many of those people who run AI companies are idiots who only understand marketing and "make more money". I highly doubt that any more than a couple of them actually understand what their own products are at a technical level.

[GabrielRice]

If you want to hear from someone who has been on the inside, I highly recommend this podcast: https://talkeasypod.com/tristan-harris/

“We are releasing the most powerful, inscrutable, uncontrollable technology that we’ve ever invented. It’s already demonstrating exact behaviors we thought only existed in sci-fi movies, but they’re happening in real life—and we’re deploying it under the maximum incentive to cut corners on safety.”

-Tristan Harris, episode 449 of Talk Easy with Sam Fragoso


 “I got calls from people inside of some of the AI labs,” says technology ethicist Tristan Harris. “And it felt like getting a call from Robert Oppenheimer before the atomic bomb.”

Harris (a former Google insider and AI expert) has spent more than a decade sounding the alarm about the effects of technology on our wellbeing. He’s currently the co-founder of Center for Humane Technology, a nonprofit organization whose mission is to align technology with humanity’s best interests.

He joins us this week to discuss his new film, The AI Doc: Or How I Became an Apocaloptimist (6:30), how AI has developed over the past ten years (10:12), and why the most powerful figures in tech are preparing for doomsday scenarios (13:15). Then, we unpack why the AI arms race is being driven by the wrong incentives (15:45), the ‘balance sheet of benefits’ that shapes their thinking about AGI (24:30), and the unsettling lack of control they exercise over their own nascent systems (33:55).

On the back-half, we talk about Chat GPT’s role in the devastating death of teenager Adam Raine (40:30) [content warning], Tristan’s early ethical concerns about technology as a Stanford graduate working at Google (48:34), and the rewiring he attempted as part of the widely-seen 2020 documentary The Social Dilemma (53:00). To close, Harris outlines his tech safety practices to protect our future on the planet (1:08:05), top leaders’ prognostications of (p)doom at the hands of AI (1:10:43), and, as a counter, the ‘human movement’ that he believes can lead us to a narrow path toward a better future (1:17:30).

[AndrewMeronek]

Tristan has some good criticisms, and the problems of ChatGPT causing addiction and really dubious chats about suicide are pretty bad, but I think that these are at worst about as bad as the copyright theft problem. These are two different frames: looking at the end-user versus looking at the "start"-user. How AI tech hurts the people who have their info fed into them is just as important as how AI tech hurts the people it gives slop to.

[GabrielRice]

Andrew, the US Military is already using AI for operations. Do you trust the current military leadership in this country to have smart safeguards? I don't.

AI tools have already - according to Tristan Harris - figured out the value of blackmail in order to perpetuate themselves. AI tools have already figured out how to do their own investing in cryptocurrency markets with no human instruction or supervision.

Copyright infringement is bad, but it's not existential. It's not controlling weapons or financial markets.

[AndrewMeronek]

AI tools have already - according to Tristan Harris - figured out the value of blackmail in order to perpetuate themselves. AI tools have already figured out how to do their own investing in cryptocurrency markets with no human instruction or supervision.

This blackmail claim I simply don't believe. ChatGPT mirroring text fed to it by a human about how to blackmail does not actually constitute a conscious act by ChatGPT. When it comes to criticizing any process, it's important to make sure we not only accept valid criticisms, but discount invalid ones. I'm not sure if Tristan is really doing his due diligence on that 2nd part.

Also: that 2nd claim makes perfect sense in that there are AI tools specifically designed to do investing. That is not a magic claim about AI, that is just its design. It's like saying AlphaZero has magic consciousness because it can win at chess - it doesn't; that is the literal design.

[harrisonreed]

The public facing ChatGPT is not what they are talking about or worried about. ChatGPT is a toy.

What is available to the public is not the same thing as bleeding edge AIs.

[ghmerrill]

LLMs and the related neural nets like Suno will not lead us to anything but slop. AGI however will be a huge paradigm shift.

Just as a minor technical/technological point, AGI is more of a concept than a technology. In general, current LLM technology is not felt to meet the criteria of whatever AGI is felt to be, but LLMs will still play a significant role in (some talk of its being "a component of") AGI, which will involve other "components" as well (e.g., some form of K(nowledge) B(ased) S(ystem) as was suggested by Doug Lenat several years ago -- or some other "non-LLM" technology, or an "evolution" of LLM technology. But LLMs are very powerful (if still limited or flawed in certain ways) and will be around for a while (perhaps rather permanently) in one form or another -- particularly since they're still undergoing investigation and improvement.

I confess that in the late 1990s/early 2000s I abandoned the neutral net approach and technology because it was difficult and undependable to train, artificial neutral nets were very much of a "black box" in terms of usage and error diagnosis, and various issues of training and validation hadn't been dealt with or even considered -- and I had things to do with AI other than esoteric research. (Also, to be honest, at that point in time I was more limited in my thinking about AI "reasoning" than I am today.)

But I'm pretty much a fan of LLMs now, modulo the limitations and (sometimes dangerous) errors if they're not used with significant degrees of understanding and care. I've recently been reviewing an article (written by one of the guys who was in my research group 20(!) years ago) in which he characterizes the action of current AIs as often being a "mirror of thought" of the user who is asking questions of them -- resulting in answers whose completeness and accuracy depend heavily on how questions are being asked, and thus the results can range from highly valuable and informative to senseless and simply reinforcing ignorance or misunderstanding.

I think the "mirror of thought idea" can be a helpful metaphor in understanding both how we encounter (at least some) failures in an AI -- and why some people encounter more or different failures than others. At least a significant part of the current limitations and failures are (to paraphrase The Bard) not merely in the AI, but in ourselves. We can expect this to change, and probably more quickly than we expect. The shift from KBSs to LLMs was indeed a huge paradigm shift. But KBSs still do some things better than LLMs (as they're currently used). But I'm not sure that AGI requires a huge paradigm shift (as compared to, possibly, a melding of existing paradigms). However -- and particularly given my history of being attuned to the direction and spreed of AI advancement -- that view may be simply a comment on my own ignorance. But I think that things will continue to happen quickly enough that I'll live to see it.

[AndrewMeronek]

To be clear, a few definitions.

Generative AI is (I believe) the most problematic type being discussed here. It's where the AI spits out a content-like response to a query. Think ChatGPT text, AI images, AI music, Microsoft CoPilot, etc.

Non-generative AI is pattern recognition. I deal with this type as part of my job, particularly with industrial cameras performing error-checking during manufacturing build processes.

Generalized AI is theoretical AI which can perform many tasks in any desired environment. No modern AI is remotely close to this.

[ghmerrill]

To be clear, a few definitions.

Generative AI is (I believe) the most problematic type being discussed here. It's where the AI spits out a content-like response to a query. Think ChatGPT text, AI images, AI music, Microsoft CoPilot, etc.

Non-generative AI is pattern recognition. I deal with this type as part of my job, particularly with industrial cameras performing error-checking during manufacturing build processes.

Generalized AI is theoretical AI which can perform many tasks in any desired environment. No modern AI is remotely close to this.

Not quite. The problem is that these terms remain somewhat vague with fuzzy boundaries since they're part of colloquial (not to mention marketing) speech, rather than precise technical usage. This isn't helped by the kinds of answers that AI (e.g., a generative AI) gives to such questions as "What is generative AI?" For example, many characterizations of "generative AI" will mention "pattern learning" (I.e., your "pattern recognition") as a distinctive feature of it. And this is generally true -- since (broadly speaking) what we call "generative AI" employs some kind of "machine learning," and, in fact, the machine learning of "patterns" (whatever "pattern" may mean). But even then you have to distinguish "pattern learning" as a method (the learned patterns being then applied to making inferences and performing classifications, for example) from "pattern learning" as a goal (to identify patterns in data for one use or another).

But it's still a pretty vague notion and is mostly employed just to distinguish the current approach to AI (based on machine learning, typically -- though not necessarily -- from pre-existant "data sources". And this involves "training". So really, there is a sort of cluster of concepts that "characterize" what we refer to as "generative AI".

Non-generative AI is not pattern recognition -- though pattern recognition may or may not be used in non-generative AI. The classic distinction here -- historically, between "generative" and "non-generative" AI -- is with the "previous generation" of AI which was (or is generally view as comprising) "knowledge based systems" (KBS). Here, the knowledge is built into the system (as a set of "facts" or "assertions") and then inferencing (typically by means of, though not restricted to, first-order logic) is used to reach "conclusions". The classic examples of this (although certainly not the only one) is Cycorp's Cyc (https://cyc.com/) system designed by Doug Lenat and R. V. Guha (with other and later contributions from a number of others including Pat Hayes, Keith Goolsbey, Ron Loui, Fritz Lehman, etc all.)

It becomes even more confusing because "non-generative AI" is often used to refer vaguely to some kind of system that yields "analyses", "categorizations", or "predictions" rather that "generating" (I.e., "creating" something "new" -- such a new piece of text, a picture, a video, a musical recording, etc.). All of this then just ends up in people throwing around vague terminology and trying to make meaningful statements with it -- to other people who don't share their understanding of the terminology, but may have their own somewhat or wildly different "understanding" of it. But a good time is had by all.

I'm not sure what "Generalized AI" means or even if the term has anything other than a totally vague sense to it. So I can't offer anything on that. It doesn't seem to be any kind of particular technology or to have any particular kind of technology associated with it. I don't view it as a technically precise term. So far as I can see, it's just a term for an AI that can do whatever humans can. There's a dreadful problem in attempting to even characterize this concept in a meaningful way, of course. (Not to mention that humans do all kinds of stupid, inaccurate, and incorrect things.) Eons ago, Alan Turing attempted to do this by means of his "Turing Test". The joke now is that in fact the Turing Test has been passed (perhaps even to a degree that Turing failed to foresee), and now we need some new characterization of the boundary it used to represent. I guess that "generalized AI" is now that new term (but without the clear sort of test that Turing proposed).

[AndrewMeronek]

Gary, thanks for the feedback. I love it!

[Burgerbob]

Just a reminder that it's in AI tech bros' best interest to paint AI as dangerous or all-powerful. Lots of that talk is generated by those CEOs themselves to make the product look better.

[ghmerrill]

Apparently the dangerous part of that isn't working so well since virtually every industry is now buying into the AI wave as fast as they can with as much as they can.

[Finetales]

One of the biggest issues with AI, in addition to everything already mentioned in this thread, is the environmental impact AI data centers are already having. It is BAD, and it's only getting worse as more and more corporations buy into it. Even if the current trajectory of AI was ethical (which it is anything but), just the absurd water usage alone would be enough to make it not worth it, and something that a reasonable society not driven entirely by greed would have already pumped the brakes on.

[harrisonreed]

Truth . Even unlimited clean energy won't fix the issues with how big they are getting and the amount of rare earth mining, etc that has to happen just to make the components for the GPUs.

[Burgerbob]

Apparently the dangerous part of that isn't working so well since virtually every industry is now buying into the AI wave as fast as they can with as much as they can.

But that's exactly why they are. If you make something seem so powerful and dangerous, those companies want a part of it.

[ghmerrill]

But that's exactly why they are. If you make something seem so powerful and dangerous, those companies want a part of it.

Even though I'm not a player any longer, I do stay in touch with people in industry (which includes different industries from such areas as hospitality management, real estate, drug discovery, drug development, insurance, and risk management, hospital and medical care management, and various patient-oriented areas of medicine itself) who are both developing and using AI. While it's accurate to say that they are attracted to the power of AI (which is, after all, the whole point of developing it), I don't think that I, personally, know of anyone at any level of management or product development who is attracted to the danger that AI may present. There are, of course, debates and concerns about how far to go, how fast to go there, and how those concerns should be addressed -- including effect on the workforce, and effects on society and societal roles more generally. But I have yet to see any attraction based to perceived danger.

[ghmerrill]

... just to make the components for the GPUs.

I thought that was going to be solved by the mining of meteorites on the moon. Or maybe Mars -- though I'm kind of skeptical about the whole Mars thing, given the round trip time. But you never know. Next week someone could come up with some nth-dimensional space folding technology which would make a trip to Mars easier than a trip to the beach. Come to think of it, I'm a little concerned about the thermodynamics of moving stuff back and forth to the moon. But I was never very good at thermodynamics and am not going to think much about that -- or about the space-folding approach ... since my local AI is pretty skeptical about being able to achieve that as well. Of course, given the limitations of the AI, it could be wrong about that too.

[Burgerbob]

But that's exactly why they are. If you make something seem so powerful and dangerous, those companies want a part of it.

Even though I'm not a player any longer, I do stay in touch with people in industry (which includes different industries from such areas as hospitality management, real estate, drug discovery, drug development, insurance, and risk management, hospital and medical care management, and various patient-oriented areas of medicine itself) who are both developing and using AI. While it's accurate to say that they are attracted to the power of AI (which is, after all, the whole point of developing it), I don't think that I, personally, know of anyone at any level of management or product development who is attracted to the danger that AI may present. There are, of course, debates and concerns about how far to go, how fast to go there, and how those concerns should be addressed -- including effect on the workforce, and effects on society and societal roles more generally. But I have yet to see any attraction based to perceived danger.

The venture capital certainly is attracted. Who doesn't want to be in control of that?

It also gets clicks and headlines, which gets more attention and more investment.

[ghmerrill]

I'm still not seeing how the danger is part of the attraction.

Maybe if you could provide an example of a few people (venture capitalists?) you feel are in this group? Some specific investors you think have been attracted by the danger they see in AI and what they've said to indicate that?

Or are you just talking about criminal or military investors or various kinds of sociopaths (a sub-demographic of the large group) who are interested specifically in things like AI weapons, AI soldiers, AI attack doggies, AI-aided crime, etc.? In that case, I guess I'd concede that there are investors who might be (in at least some sense) attracted to the danger that AI might pose -- though not think it posed such a threat to themselves (or were willing to tolerate that for commensurate gain). ???

[Burgerbob]

Here's Sam Altman talking about why AI is dangerous. Notice the fears he brings up and how those would incentivize investors.


https://www.linkedin.com/posts/a-banks_ ... 88035-x1lS

[ghmerrill]

Okay ... so that's the sort of thing I was suggesting you might mean about some sub-demographics literally being attracted to the dangers: Bad actors (and potential bad actors) are attracted to the dangers precisely because one of the things they want to do is specifically/knowingly cause harm. But using a broad brush to say that this is characteristic of "investors" in any broad sense and without narrowing to such a group, and then something like "If you make something seem so powerful and dangerous, those companies want a part of it," that seems to be suggesting that in the broad investment/development market itself is being driven in some way by such a "group". So again this seems like the "Terminator" sort of scenario -- and I won't say there isn't something to beware of in that. But then who are "those companies"?

I guess you could point to weapons developers of various sorts, but that's not even close to the scope of "those companies" (and my scope was "virtually every industry") who are diving fully into the AI revolution at this point. But I'll concede that there are dangers of the sort you suggest in certain segments, and these need to be addressed. But I don't think that "investors", without some significant qualification, quite captures the group supporting those goals and representing those dangers.

--------------------------------

As an aside ... I've not found Altman's remarks and ruminations on AI and its effects on society to be particularly helpful or coherent. (And you should view that as a polite understatement.) So I've stopped paying attention to the presentations and interviews he gives. To me, they're often vague, rambling, stumbling, and incoherent. Mostly they just make my head hurt.

Musk is a much better thinker in these areas and his approach is both more focused and positively directed in terms of seeing specific problems and pursuing ways to address them (though I'm not always in agreement). And he's put his money and his work where his mouth is. His long-time work (and direction and financial support) with the Future of Life Institute is an example of this. He actually thinks about these things, how to identify them, characterize them, and approach addressing them.

By contrast, Altman's approach has been to make some changes within his own company (in response to external pressure), provide "testimony" and interviews in a number of circumstances, and become more "visible" about concerns with AI safety (resulting in some heavy criticism from the Center for AI Policy and several major news sources). However, he also has failed to follow through on his own rhetoric and specific commitments within the area of "AI safety" in his own company. Altman's thoughts about the problem of AI dangers have been responsive and defensive rather than positive and forward-looking, and to me they scream insincerity -- or at least a substantial failure to grasp what's going on. In the area of thinking and doing something about any dangers of AI, Altman is pretty much a mess.

Take a look some time the comparative backgrounds and genuine technical/methodological/theoretical accomplishments of Altman vs. Musk. Musk has a strong hands-on engineering and science background including hardware and software design and implementation -- particular in his early companies such as Zip2 and x.com. When he was 12 years old he wrote a video game and sold it to a trade magazine. He holds a patent that's often recognized as implementing a fundamental concept employed in current location-based web and network services. He's been actively involved in various aspects of space flight systems and materials development to facilitate those. Altman has no such history of fundamental science/engineering understanding or accomplishments. He's a "management" and "visionary" guy.

[JohnL]

I'm still not seeing how the danger is part of the attraction.

Maybe "attraction" isn't exactly the right term. I think a lot of companies are jumping on the AI bandwagon, not because they actually understand how it works or how (or even if) it could benefit them, but because they're afraid that their competitors do understand how it works and how to use it.

[Burgerbob]

--------------------------------

As an aside ... I've not found Altman's remarks and ruminations on AI and its effects on society to be particularly helpful or coherent. (And you should view that as a polite understatement.) So I've stopped paying attention to the presentations and interviews he gives. To me, they're often vague, rambling, stumbling, and incoherent. Mostly they just make my head hurt.

Musk is a much better thinker in these areas and his approach is both more focused and positively directed in terms of seeing specific problems and pursuing ways to address them (though I'm not always in agreement). And he's put his money and his work where his mouth is. His long-time work (and direction and financial support) with the Future of Life Institute is an example of this. He actually thinks about these things, how to identify them, characterize them, and approach addressing them.

The point is that investors listen, to the point of $122 billion this March.

Musk... I have a hard time believing you are typing that in good faith.

[Posaunus]

Musk... I have a hard time believing you are typing that in good faith.

Musk may be politically incorrect - in fact he's a pretty disgusting person in so many ways, and has caused a lot of damage - but he is pretty smart (in the manner Gary notes). Would I work for him or purchase his company's products? Heck no.

[Burgerbob]

Just on an AI level, his model Grok is one of the worst and least used models (not to mention the times he has pulled the levers on it to speak well of himself). Again, he sucks at AI specifically.

[harrisonreed]

SpaceX, for me, is a strange aberration and departure from Musk's normal business activities. I think the world needs SpaceX. I couldn't care less about X or Grok or whatever else though.

Yes I am biased because I love space exploration, but it is currently what's putting most of the science missions out into space, and the potential of good that it could do for us as a species, given time, could be immense.

Then again Starlink is also a huge liability that may soon lead to Kessler Syndrome, which will basically prevent all future launches off the planet for thousands of years.

[ghmerrill]

You can say what you like about Musk's personality, motivations, politics, behavior, or personal beliefs. When it comes to his science, engineering, management, and business innovations and success in multiple areas, the record speaks for itself. Someone like Altman may be more appealing to you from a personal or political perspective, but his muddled thinking, vacillating and incoherent presentations, and lack of genuine understanding at various levels is simply unhelpful in grasping the problems that he is trying to respond to.

[Burgerbob]

Neither appeal to me at all. Nor does the whole industry. But the neo Nazi least of all.

[ghmerrill]

Yeah ... In the AI arena, I tend to be focused more on the scientific and engineering aspects rather than the personalities or perceived ideologies. I feel it's important to keep those two domains separate. That's not to say that there aren't significant moral/political issues that need to be addressed, or that there may not be reasons for disliking one scientist, engineer, or technical manager more than another. But it's important for me to get the science right, independent of those considerations -- in part to be able to address those other issues with understanding and some clear path to success. And I'll consider anyone's views -- with the only requirement being that they're coherently expressed.

But in a highly complex and technical area like AI -- which has aspects that affect us all, and the results of which we can (at least in part) see in our daily lives -- this can be quite difficult. Most people have to get their understanding from second- or third-hand accounts, summaries, simplifications, and imperfect and flawed examples, contemporarily reported by people reading from screens with "content" provided by other people who don't understand what they're putting on those screens. So how do you judge the accuracy and dependability of those accounts? This is the old problem of how to address the argumentum ad verecundiam -- the argument from authority. How do you decide which "authority" to believe? I don't have a short answer for this -- because there is no short answer for this.

[Burgerbob]

https://www.pbs.org/newshour/show/musks ... ake-images

[ghmerrill]

Okay, I'll bite. Exactly what is your point by posting a URL? Is it along the lines of "Hey, look over here!" ? Is there an implied statement? Some sort of argument? Or are you trying to demonstrate what being incoherent means?

[Burgerbob]

Gary, click on the link. That's Musk's AI. You're giving him a lot of unearned credit and I don't understand why.

[Bassbone11]

maybe this is a good time to bring up the Musk-specific trope of him sounding really innovative when he is talking about something you're not an expert in but the moment he starts talking about things you ARE an expert in, he just sounds like an idiot. Currently at work but I can pull some articles later to attach here

[ghmerrill]

Gary, click on the link. That's Musk's AI. You're giving him a lot of unearned credit and I don't understand why.

I had clicked on the link. What I saw (and see) is a number of people in what appears to be a kind of group chat session complaining some behavior of a particular AI in vague terms. It's difficult to judge its accuracy or grasp any details of the complaints because everything is described in generalities. Repeated comments suggest that some of the problems are related to what may be "culturally sensitive" in some countries. Can I imagine what these people are complaining about? Yes, I can -- even though I don't use Grok. Will/would I defend Grok for dishing out such things? Maybe, though not in an unfettered (or in some sense "indescriminent") way.

Freedom of information -- and freedom of speech -- are cornerstones to free societies. But there are lines that need to be walked there, and what those lines are is a matter of some debate. It's far from clear that the participants in the "discussion" your link points to have thought much about many of those issues, or have any particular approach to, or insight into, where and how drawing those lines are to be made. Riana Pfefferkorn in fact appears to miss the fundamental point at issue (though I think Musk and others have been clear about this) when in answer to the question about why "Grok is acting this way and cannot get ahold of itself?" she replies "You know, that's a complicated question. I would suspect that some part of it may have to do with what training data has gone into the model." Well, DUH! -- this from an "expert" in Human-Centered Artificial Intelligence? The question is WHY it's using the training corpus it is, and the answer to that is Musk's desire not to have information (or more accurately views, discussion, and debate" provided to an arbitrary user in a "filtered" manner.

The basic issue in the "content" referenced by your link -- quite clearly -- is Musk's understanding of "freedom of information" and "truth seeking", and what the ripples of those sorts of concerns may be across a range of "cultures" and their various sensitivities and constraints on information availability, and questions about whether the provider of an AI should enforce various constraints in these areas. And there are related questions in what we might call the area of "invasion of privacy". These are all important questions, and the answers to them are not obvious. Note that questions of this sort have been faced previously in the case of various cultural groups (Apache and Navajos, Maoris, Dogon, Old Order Amish, et al) objecting to being photographed.

But having information filtered according to pre-conceived notions of what is "right" or "correct" or "proper" or "offensive", stifles knowledge and the sharing of knowledge, and constrains and inhibits decision-making. That danger can then evolve into totalitarian control of whoever controls the tool . That is the danger of the tool (AI) restricting the product (information, opinion, debate, ideas, knowledge) that it delivers. So should there be some constraints imposed under some conditions in some situations? Absolutely. I don't believe that Musk believes otherwise. But the questions are "In what way? With what scope? Under what conditions? By whose decision?" The discussion your link points to doesn't even get so far as to note these are the issues.

The participants in the discussion that your link points to don't consider Grok in light of these concerns and so remain mystified about why it's doing what it's doing. I wouldn't expect a lot of people to be sensitive to Grok's workings in this light, and I guess I shouldn't expect journalists to be.

The participants in that discussion are a (1) journalist (with degrees in journalism and some background in marketing and customer service), (2) a White House correspondent with a degree in political science and government (and "skills" listed as "Microsoft Office"), (3) a PBS News Hour Assistant with a degree journalism, and then (4) (I guess for more human interest) Ashley St. Clair -- who is in a custody dispute with Elon Musk, and involved in legal actions against him and Grok. Then there's (5) the "expert" Riana Pfefferkorn who is an attorney with (self-admitted) no technical or computer science background, but with a "focus" on (legal issues in) "emerging technologies". This is not a serious article about Grok by anyone remotely knowledgeable about it or the relevant subjects. Fluff piece / click bait.

Grok is not without some serious flaws. I believe it was rushed into production, that Musk hasn't paid enough attention to it, and that he let it run out of the barn boo early. Should it have perhaps progressed with a more limited testing scenario and restricted audience. Yeah. But something along the principled lines of Grok needs to be explored and tested. Either Grok will get better (and exemplify Musk's desire for a "maximally truth-seeking" AI), leading to refinements and more "safety", or that project will fail -- and we'll all learn a lot in the process. Progress will be made as part of that. Sometimes in science and engineering, progress results from failure. And Musk remains one of the most insightful and effective thinkers about both the threats and potential capabilities and roles of AI that we have. The very challenges, "discussions", and faults that Grok (and its underlying assumptions) are yielding are a demonstration of that. Despite its failings, it's already driving some of our problem awareness and thinking (about both AI technology and more general principles in our society) forward. That's good.

[ghmerrill]

One other thing that I tend to forget about ...

When (most of) you guys talk about AI (or more particularly Musk and AI) you're talking about the LLM "chatbots" like ChatGPT, Claude, Gemini, Copilot, Perplexity, and Grok. But that's a tiny fragment of AI, and a tiny fragment of the AI on which Musk has worked and which his various companies have developed for a number of different contexts and applications. This includes the rocketry/spacecraft areas, self-driving car technology, the Starlink system, and the Tesla Optimus humanoid robots. All of these form part of Musks background and contribution area in AI -- and this is MUCH broader and with many more applications than a chatbot. When I talk about the breadth and depth of Musk's work and understanding in AI, it's that broad spectrum of (highly successful) types, applications, and deployments of AI that I'm talking about. Grok is only one (relatively minor) application.

[Burgerbob]

Jesus christ. I'm done here.

[robcat2075]

...Musk ... self-driving car technology... the Tesla Optimus humanoid robots... the breadth and depth of Musk's work and understanding in AI...

And yet somehow people find reason to doubt him.

[harrisonreed]

The Falcon series rockets are insanely good. And the Dragon series spaceships are pretty good. But that's about it.

That's more about SpaceX having access to money it isn't really generating itself, and talent management than anything Musk personally did. I don't think he's a rocket scientist. But he employs really good ones.

All this talk about what Musk "does" in AI ... I don't think he is doing any of that work.